Commit dd0107a0 authored by Sagi Grimberg's avatar Sagi Grimberg Committed by Doug Ledford

IB/iser: set block queue_virt_boundary

The block layer can reliably guarantee that SG lists won't
contain gaps (page unaligned) if a driver set the queue
virt_boundary.

With this setting the block layer will:
- refuse merges if bios are not aligned to the virtual boundary
- split bios/requests that are not aligned to the virtual boundary
- or, bounce buffer SG_IOs that are not aligned to the virtual boundary

Since iser is working in 4K page size, set the virt_boundary to
4K pages. With this setting, we can now safely remove the bounce
buffering logic in iser.
Signed-off-by: default avatarSagi Grimberg <sagig@mellanox.com>
Reviewed-by: default avatarChristoph Hellwig <hch@lst.de>
Signed-off-by: default avatarDoug Ledford <dledford@redhat.com>
parent 6c760b3d
...@@ -762,9 +762,7 @@ iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *s ...@@ -762,9 +762,7 @@ iscsi_iser_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *s
stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */ stats->r2t_pdus = conn->r2t_pdus_cnt; /* always 0 */
stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt; stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt; stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
stats->custom_length = 1; stats->custom_length = 0;
strcpy(stats->custom[0].desc, "fmr_unalign_cnt");
stats->custom[0].value = conn->fmr_unalign_cnt;
} }
static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep, static int iscsi_iser_get_ep_param(struct iscsi_endpoint *ep,
...@@ -969,6 +967,13 @@ static umode_t iser_attr_is_visible(int param_type, int param) ...@@ -969,6 +967,13 @@ static umode_t iser_attr_is_visible(int param_type, int param)
return 0; return 0;
} }
static int iscsi_iser_slave_alloc(struct scsi_device *sdev)
{
blk_queue_virt_boundary(sdev->request_queue, ~MASK_4K);
return 0;
}
static struct scsi_host_template iscsi_iser_sht = { static struct scsi_host_template iscsi_iser_sht = {
.module = THIS_MODULE, .module = THIS_MODULE,
.name = "iSCSI Initiator over iSER", .name = "iSCSI Initiator over iSER",
...@@ -982,6 +987,7 @@ static struct scsi_host_template iscsi_iser_sht = { ...@@ -982,6 +987,7 @@ static struct scsi_host_template iscsi_iser_sht = {
.eh_target_reset_handler = iscsi_eh_recover_target, .eh_target_reset_handler = iscsi_eh_recover_target,
.target_alloc = iscsi_target_alloc, .target_alloc = iscsi_target_alloc,
.use_clustering = DISABLE_CLUSTERING, .use_clustering = DISABLE_CLUSTERING,
.slave_alloc = iscsi_iser_slave_alloc,
.proc_name = "iscsi_iser", .proc_name = "iscsi_iser",
.this_id = -1, .this_id = -1,
.track_queue_depth = 1, .track_queue_depth = 1,
......
...@@ -227,18 +227,13 @@ enum iser_data_dir { ...@@ -227,18 +227,13 @@ enum iser_data_dir {
* @size: num entries of this sg * @size: num entries of this sg
* @data_len: total beffer byte len * @data_len: total beffer byte len
* @dma_nents: returned by dma_map_sg * @dma_nents: returned by dma_map_sg
* @orig_sg: pointer to the original sg list (in case
* we used a copy)
* @orig_size: num entris of orig sg list
*/ */
struct iser_data_buf { struct iser_data_buf {
struct scatterlist *sg; struct scatterlist *sg;
unsigned int size; unsigned int size;
unsigned long data_len; unsigned long data_len;
unsigned int dma_nents; unsigned int dma_nents;
struct scatterlist *orig_sg; };
unsigned int orig_size;
};
/* fwd declarations */ /* fwd declarations */
struct iser_device; struct iser_device;
......
...@@ -661,48 +661,14 @@ void iser_task_rdma_init(struct iscsi_iser_task *iser_task) ...@@ -661,48 +661,14 @@ void iser_task_rdma_init(struct iscsi_iser_task *iser_task)
void iser_task_rdma_finalize(struct iscsi_iser_task *iser_task) void iser_task_rdma_finalize(struct iscsi_iser_task *iser_task)
{ {
int is_rdma_data_aligned = 1;
int is_rdma_prot_aligned = 1;
int prot_count = scsi_prot_sg_count(iser_task->sc); int prot_count = scsi_prot_sg_count(iser_task->sc);
/* if we were reading, copy back to unaligned sglist,
* anyway dma_unmap and free the copy
*/
if (iser_task->data[ISER_DIR_IN].orig_sg) {
is_rdma_data_aligned = 0;
iser_finalize_rdma_unaligned_sg(iser_task,
&iser_task->data[ISER_DIR_IN],
ISER_DIR_IN);
}
if (iser_task->data[ISER_DIR_OUT].orig_sg) {
is_rdma_data_aligned = 0;
iser_finalize_rdma_unaligned_sg(iser_task,
&iser_task->data[ISER_DIR_OUT],
ISER_DIR_OUT);
}
if (iser_task->prot[ISER_DIR_IN].orig_sg) {
is_rdma_prot_aligned = 0;
iser_finalize_rdma_unaligned_sg(iser_task,
&iser_task->prot[ISER_DIR_IN],
ISER_DIR_IN);
}
if (iser_task->prot[ISER_DIR_OUT].orig_sg) {
is_rdma_prot_aligned = 0;
iser_finalize_rdma_unaligned_sg(iser_task,
&iser_task->prot[ISER_DIR_OUT],
ISER_DIR_OUT);
}
if (iser_task->dir[ISER_DIR_IN]) { if (iser_task->dir[ISER_DIR_IN]) {
iser_unreg_rdma_mem(iser_task, ISER_DIR_IN); iser_unreg_rdma_mem(iser_task, ISER_DIR_IN);
if (is_rdma_data_aligned)
iser_dma_unmap_task_data(iser_task, iser_dma_unmap_task_data(iser_task,
&iser_task->data[ISER_DIR_IN], &iser_task->data[ISER_DIR_IN],
DMA_FROM_DEVICE); DMA_FROM_DEVICE);
if (prot_count && is_rdma_prot_aligned) if (prot_count)
iser_dma_unmap_task_data(iser_task, iser_dma_unmap_task_data(iser_task,
&iser_task->prot[ISER_DIR_IN], &iser_task->prot[ISER_DIR_IN],
DMA_FROM_DEVICE); DMA_FROM_DEVICE);
...@@ -710,11 +676,10 @@ void iser_task_rdma_finalize(struct iscsi_iser_task *iser_task) ...@@ -710,11 +676,10 @@ void iser_task_rdma_finalize(struct iscsi_iser_task *iser_task)
if (iser_task->dir[ISER_DIR_OUT]) { if (iser_task->dir[ISER_DIR_OUT]) {
iser_unreg_rdma_mem(iser_task, ISER_DIR_OUT); iser_unreg_rdma_mem(iser_task, ISER_DIR_OUT);
if (is_rdma_data_aligned)
iser_dma_unmap_task_data(iser_task, iser_dma_unmap_task_data(iser_task,
&iser_task->data[ISER_DIR_OUT], &iser_task->data[ISER_DIR_OUT],
DMA_TO_DEVICE); DMA_TO_DEVICE);
if (prot_count && is_rdma_prot_aligned) if (prot_count)
iser_dma_unmap_task_data(iser_task, iser_dma_unmap_task_data(iser_task,
&iser_task->prot[ISER_DIR_OUT], &iser_task->prot[ISER_DIR_OUT],
DMA_TO_DEVICE); DMA_TO_DEVICE);
......
...@@ -88,113 +88,6 @@ int iser_assign_reg_ops(struct iser_device *device) ...@@ -88,113 +88,6 @@ int iser_assign_reg_ops(struct iser_device *device)
return 0; return 0;
} }
static void
iser_free_bounce_sg(struct iser_data_buf *data)
{
struct scatterlist *sg;
int count;
for_each_sg(data->sg, sg, data->size, count)
__free_page(sg_page(sg));
kfree(data->sg);
data->sg = data->orig_sg;
data->size = data->orig_size;
data->orig_sg = NULL;
data->orig_size = 0;
}
static int
iser_alloc_bounce_sg(struct iser_data_buf *data)
{
struct scatterlist *sg;
struct page *page;
unsigned long length = data->data_len;
int i = 0, nents = DIV_ROUND_UP(length, PAGE_SIZE);
sg = kcalloc(nents, sizeof(*sg), GFP_ATOMIC);
if (!sg)
goto err;
sg_init_table(sg, nents);
while (length) {
u32 page_len = min_t(u32, length, PAGE_SIZE);
page = alloc_page(GFP_ATOMIC);
if (!page)
goto err;
sg_set_page(&sg[i], page, page_len, 0);
length -= page_len;
i++;
}
data->orig_sg = data->sg;
data->orig_size = data->size;
data->sg = sg;
data->size = nents;
return 0;
err:
for (; i > 0; i--)
__free_page(sg_page(&sg[i - 1]));
kfree(sg);
return -ENOMEM;
}
static void
iser_copy_bounce(struct iser_data_buf *data, bool to_buffer)
{
struct scatterlist *osg, *bsg = data->sg;
void *oaddr, *baddr;
unsigned int left = data->data_len;
unsigned int bsg_off = 0;
int i;
for_each_sg(data->orig_sg, osg, data->orig_size, i) {
unsigned int copy_len, osg_off = 0;
oaddr = kmap_atomic(sg_page(osg)) + osg->offset;
copy_len = min(left, osg->length);
while (copy_len) {
unsigned int len = min(copy_len, bsg->length - bsg_off);
baddr = kmap_atomic(sg_page(bsg)) + bsg->offset;
if (to_buffer)
memcpy(baddr + bsg_off, oaddr + osg_off, len);
else
memcpy(oaddr + osg_off, baddr + bsg_off, len);
kunmap_atomic(baddr - bsg->offset);
osg_off += len;
bsg_off += len;
copy_len -= len;
if (bsg_off >= bsg->length) {
bsg = sg_next(bsg);
bsg_off = 0;
}
}
kunmap_atomic(oaddr - osg->offset);
left -= osg_off;
}
}
static inline void
iser_copy_from_bounce(struct iser_data_buf *data)
{
iser_copy_bounce(data, false);
}
static inline void
iser_copy_to_bounce(struct iser_data_buf *data)
{
iser_copy_bounce(data, true);
}
struct iser_fr_desc * struct iser_fr_desc *
iser_reg_desc_get_fr(struct ib_conn *ib_conn) iser_reg_desc_get_fr(struct ib_conn *ib_conn)
{ {
...@@ -238,62 +131,6 @@ iser_reg_desc_put_fmr(struct ib_conn *ib_conn, ...@@ -238,62 +131,6 @@ iser_reg_desc_put_fmr(struct ib_conn *ib_conn,
{ {
} }
/**
* iser_start_rdma_unaligned_sg
*/
static int iser_start_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
struct iser_data_buf *data,
enum iser_data_dir cmd_dir)
{
struct ib_device *dev = iser_task->iser_conn->ib_conn.device->ib_device;
int rc;
rc = iser_alloc_bounce_sg(data);
if (rc) {
iser_err("Failed to allocate bounce for data len %lu\n",
data->data_len);
return rc;
}
if (cmd_dir == ISER_DIR_OUT)
iser_copy_to_bounce(data);
data->dma_nents = ib_dma_map_sg(dev, data->sg, data->size,
(cmd_dir == ISER_DIR_OUT) ?
DMA_TO_DEVICE : DMA_FROM_DEVICE);
if (!data->dma_nents) {
iser_err("Got dma_nents %d, something went wrong...\n",
data->dma_nents);
rc = -ENOMEM;
goto err;
}
return 0;
err:
iser_free_bounce_sg(data);
return rc;
}
/**
* iser_finalize_rdma_unaligned_sg
*/
void iser_finalize_rdma_unaligned_sg(struct iscsi_iser_task *iser_task,
struct iser_data_buf *data,
enum iser_data_dir cmd_dir)
{
struct ib_device *dev = iser_task->iser_conn->ib_conn.device->ib_device;
ib_dma_unmap_sg(dev, data->sg, data->size,
(cmd_dir == ISER_DIR_OUT) ?
DMA_TO_DEVICE : DMA_FROM_DEVICE);
if (cmd_dir == ISER_DIR_IN)
iser_copy_from_bounce(data);
iser_free_bounce_sg(data);
}
#define IS_4K_ALIGNED(addr) ((((unsigned long)addr) & ~MASK_4K) == 0) #define IS_4K_ALIGNED(addr) ((((unsigned long)addr) & ~MASK_4K) == 0)
/** /**
...@@ -355,64 +192,6 @@ static int iser_sg_to_page_vec(struct iser_data_buf *data, ...@@ -355,64 +192,6 @@ static int iser_sg_to_page_vec(struct iser_data_buf *data,
return cur_page; return cur_page;
} }
/**
* iser_data_buf_aligned_len - Tries to determine the maximal correctly aligned
* for RDMA sub-list of a scatter-gather list of memory buffers, and returns
* the number of entries which are aligned correctly. Supports the case where
* consecutive SG elements are actually fragments of the same physcial page.
*/
static int iser_data_buf_aligned_len(struct iser_data_buf *data,
struct ib_device *ibdev,
unsigned sg_tablesize)
{
struct scatterlist *sg, *sgl, *next_sg = NULL;
u64 start_addr, end_addr;
int i, ret_len, start_check = 0;
if (data->dma_nents == 1)
return 1;
sgl = data->sg;
start_addr = ib_sg_dma_address(ibdev, sgl);
if (unlikely(sgl[0].offset &&
data->data_len >= sg_tablesize * PAGE_SIZE)) {
iser_dbg("can't register length %lx with offset %x "
"fall to bounce buffer\n", data->data_len,
sgl[0].offset);
return 0;
}
for_each_sg(sgl, sg, data->dma_nents, i) {
if (start_check && !IS_4K_ALIGNED(start_addr))
break;
next_sg = sg_next(sg);
if (!next_sg)
break;
end_addr = start_addr + ib_sg_dma_len(ibdev, sg);
start_addr = ib_sg_dma_address(ibdev, next_sg);
if (end_addr == start_addr) {
start_check = 0;
continue;
} else
start_check = 1;
if (!IS_4K_ALIGNED(end_addr))
break;
}
ret_len = (next_sg) ? i : i+1;
if (unlikely(ret_len != data->dma_nents))
iser_warn("rdma alignment violation (%d/%d aligned)\n",
ret_len, data->dma_nents);
return ret_len;
}
static void iser_data_buf_dump(struct iser_data_buf *data, static void iser_data_buf_dump(struct iser_data_buf *data,
struct ib_device *ibdev) struct ib_device *ibdev)
{ {
...@@ -483,31 +262,6 @@ iser_reg_dma(struct iser_device *device, struct iser_data_buf *mem, ...@@ -483,31 +262,6 @@ iser_reg_dma(struct iser_device *device, struct iser_data_buf *mem,
return 0; return 0;
} }
static int fall_to_bounce_buf(struct iscsi_iser_task *iser_task,
struct iser_data_buf *mem,
enum iser_data_dir cmd_dir)
{
struct iscsi_conn *iscsi_conn = iser_task->iser_conn->iscsi_conn;
struct iser_device *device = iser_task->iser_conn->ib_conn.device;
iscsi_conn->fmr_unalign_cnt++;
if (iser_debug_level > 0)
iser_data_buf_dump(mem, device->ib_device);
/* unmap the command data before accessing it */
iser_dma_unmap_task_data(iser_task, mem,
(cmd_dir == ISER_DIR_OUT) ?
DMA_TO_DEVICE : DMA_FROM_DEVICE);
/* allocate copy buf, if we are writing, copy the */
/* unaligned scatterlist, dma map the copy */
if (iser_start_rdma_unaligned_sg(iser_task, mem, cmd_dir) != 0)
return -ENOMEM;
return 0;
}
/** /**
* iser_reg_page_vec - Register physical memory * iser_reg_page_vec - Register physical memory
* *
...@@ -779,26 +533,6 @@ static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task, ...@@ -779,26 +533,6 @@ static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task,
return 0; return 0;
} }
static int
iser_handle_unaligned_buf(struct iscsi_iser_task *task,
struct iser_data_buf *mem,
enum iser_data_dir dir)
{
struct iser_conn *iser_conn = task->iser_conn;
struct iser_device *device = iser_conn->ib_conn.device;
int err, aligned_len;
aligned_len = iser_data_buf_aligned_len(mem, device->ib_device,
iser_conn->scsi_sg_tablesize);
if (aligned_len != mem->dma_nents) {
err = fall_to_bounce_buf(task, mem, dir);
if (err)
return err;
}
return 0;
}
static int static int
iser_reg_prot_sg(struct iscsi_iser_task *task, iser_reg_prot_sg(struct iscsi_iser_task *task,
struct iser_data_buf *mem, struct iser_data_buf *mem,
...@@ -841,10 +575,6 @@ int iser_reg_rdma_mem(struct iscsi_iser_task *task, ...@@ -841,10 +575,6 @@ int iser_reg_rdma_mem(struct iscsi_iser_task *task,
bool use_dma_key; bool use_dma_key;
int err; int err;
err = iser_handle_unaligned_buf(task, mem, dir);
if (unlikely(err))
return err;
use_dma_key = (mem->dma_nents == 1 && !iser_always_reg && use_dma_key = (mem->dma_nents == 1 && !iser_always_reg &&
scsi_get_prot_op(task->sc) == SCSI_PROT_NORMAL); scsi_get_prot_op(task->sc) == SCSI_PROT_NORMAL);
...@@ -867,10 +597,6 @@ int iser_reg_rdma_mem(struct iscsi_iser_task *task, ...@@ -867,10 +597,6 @@ int iser_reg_rdma_mem(struct iscsi_iser_task *task,
if (scsi_prot_sg_count(task->sc)) { if (scsi_prot_sg_count(task->sc)) {
mem = &task->prot[dir]; mem = &task->prot[dir];
err = iser_handle_unaligned_buf(task, mem, dir);
if (unlikely(err))
goto err_reg;
err = iser_reg_prot_sg(task, mem, desc, err = iser_reg_prot_sg(task, mem, desc,
use_dma_key, prot_reg); use_dma_key, prot_reg);
if (unlikely(err)) if (unlikely(err))
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment